Knowledge-Based Models of Paediatric Heart Diseases
1. Introduction
During clinical evaluation, cardiologists and clinicians measure several key parameters to identify the pathologies and to assess their severity. Ventricle volumes, contractility and blood flows for example are essential features of the cardiac function, and abnormalities in their values can be caused by heart diseases.
One of the main purposes of Health-e-Child is to gather the knowledge and the experience of the cardiologists involved in the project and to design a decision-support model that can help the community to detect and assess paediatric heart pathologies. This section is the knowledge-based model that results from the analysis of the methods used by cardiologists during the diagnostic process. It is based on the interpretation and the classification of some cardiac parameters by means of left-/right-ventricular comparisons and basic group analyses.
2. Estimation of cardiac parameters
Ultrasonography and MR imaging allow the clinicians to measure geometrical and functional parameters of the heart. From these data, they can estimate for example the volume and mass of either the whole heart or segments of interest, the blood flows, etc. More subtle measurements such as the global or segmental contractility of the myocardium can also be derived from these images, provided that 2D+t or 3D+t acquisitions are available.
In Health-e-Child, the volume of the cavities, the muscle contractility and the blood flows across the cardiac valves are estimated from 2-D/3-D ultrasonography and cardiac MR imaging.
To acquire the total volume of the right ventricle, advanced 3-D echocardiography sequences tailored to the clinical evaluation of the right ventricle (RV) in children are used (Gaslini). On the other hand, because such modality is not available for the evaluation of the left ventricle, the total volume of this cavity is estimated by using MR imaging only. Of course, when it is possible, RV volume is also measured through MRI, improving in this way previous measurements performed with ultrasonography. In both cases, the recommendation of the American Society of Echocardiography are complied.
Segmental volumes are also estimated to detect local abnormalities such as anomalous wall motions. Right ventricle for example is commonly subdivided into three segments: inlet, outlet and apical-trabecular (see next video).
|
Get the Flash Player to see this player.
|
|
Volumetric representation of the right ventricle. Yellow: outlet, Green: inlet, Red: apical. (Video from Gaslini)
|
The estimation of the volumes of the heart cavities from MR imaging is performed by using the modified Simpson's algorithm. Contrary to 2-D ultrasonography evaluations and to other algorithms, the Simpson's method does not rely on any assumptions about the heart geometry. The ventricles are first divided into 12-15 slices of approximately 8mm. The areas of the cavities are then measured at each slice and, by multiplying them by the slice thickness, the intermediate "truncated" volumes are obtained. The total volume is finally calculated by summing all the intermediate results. Owing to this algorithm and to a better image resolution, MRI clinical evaluations are often considered more reliable and more accurate.
The second set of parameters consists of blood flow measurements. During the cardiac cycle, the blood can go from the atrium to the ventricle during diastole and from the ventricle to the artery during systole. The estimation of the amount of blood that crosses the valves can thus provide useful information about the heart function. Anomalous values may point out pathological features of the myocardium, ventricular overload or valvar regurgitations. For instance, tricuspid flows that are too high with respect to the mitral flows may suggest the presence of shunt between the atria.
Flow measurements are thus crucial for diagnosis. To estimate them, ultrasonography imaging is usually used and in particular the Doppler modality. This technology relies on the Doppler effect to estimate the velocity field of the observed structures.
Concretely, blood flows across the valves are assessed as follows. The specialist first estimates the area of the valve by using ultrasonography. Then, he gets the flow velocity at the level of the valve of interest, during several cardiac cycles, by means of continuous-wave Doppler acquisitions. Finally, he obtains the mean flow that passes across this valve by using the following formula:
The following video illustrates the Doppler acquisition process. At each time t, the expert can visualise the blood velocity and the 2D plane under study. The flow that crosses the white dot line in the US image corresponds to the area of the region bounded by the velocity curve and the x-axis.
|
Get the Flash Player to see this player.
|
|
Example of Doppler data. (Video from Gaslini)
|
The last parameters that are acquired for the knowledge-based models are contractility measurements. They quantify the efficiency of the cardiac muscle. As for the previous data, those parameters are mainly obtained by using ultrasonography images and, if available, from cine MRI (3D+t).
|
Get the Flash Player to see this player.
|
|
Example of cine MRI: short axis view of the heart. (Video from Gaslini)
|
A global contractility index is measured by considering the entire ventricle, while segmental contractilities can be acquired for each segment of the heart. The ejection fraction of each ventricle (the amount of blood in percentage ejected during a cardiac cycle) is also estimated to evaluate the efficiency of the organ.
To conclude, all the clinical measurements that can be used in the knowledge-based model are summarised in the following table.
| US images | MR image |
|
|
|
|
|
3. Pre-processing of the acquired data
Once acquired, the raw clinical parameters must be pre-processed before doing any group analyses or comparisons. In this way, the values are first normalised, followed by the calculation of standard scores to point out significant differences between the patient and a population of reference.
The first step of the pre-processing stage aims at normalising the acquired values in order to remove the possible effects of the height and weight of the patient. Indeed, height and weight can have a great influence upon the cardiac function and the acquired data must therefore be normalised to cancel their impact.
In Health-e-Child, the clinical measurements are normalised by dividing them by the body surface index computed by using Dubois' equation:
After normalisation, standard scores (also known as Z-score) are computed. The standard score is a dimensionless quantity that reveals how many units of the standard deviation the patient's parameters differ from the normal values (estimated among a population of normal subjects). It is calculated by using the formula:
| zparam = | patient value - mean value |
| standard deviation |
Values higher or lower than two standard deviations (z-score higher or lower than 2) are considered as significant and may point out abnormal cardiac function.
4. Simplified Description of the Clinical Workflow
The imaging data presented in the previous section are integrated and used as input to a decision process. An exemplar system could be: the parameters are compared with normal values and displayed to the cardiologist in a user-friendly way; ventricular segments with abnormally high or low contractility are highlighted and measurements greater or lower than 2 standard deviations are pointed out; finally, Qp/Qs ratio is displayed.
In addition to the direct visualisation of the acquired parameters, right-ventricle measurements are compared with those of the left ventricle. This comparison relies on the fact that the circulatory system is a closed system and therefore the ratio of any parameter of the right ventricle over the equivalent parameter of the left ventricle must be equal to 1±5%. An anomaly in these ratios may be an indicator of pathological features.
First step: flow comparisons
The flow through the four cardiac valves will be investigated both in systole and diastole, resulting in the following eight measurements:
- Systolic and diastolic flow through the tricuspid valve
- Systolic and diastolic flow through the mitral valve
- Systolic and diastolic flow through the aortic valve
- Systolic and diastolic flow through the pulmonary valve
All the values are then integrated and a more reliable hemodynamic analysis is derived.
Broadly speaking, the blood flows could be analysed by considering the two following ratios: the atrio-ventricular ratio defined as
| RTMF = | Flow through the tricuspid valve |
| Flow through the mitral valve |
and the ventricular-arterial ratio:
| RPAF = | Flow through the pulmonary valve |
| Flow through the aortic valve |
Common clinical situations can therefore be described by comparing these two ratios, as presented in the following table:
| Value of the ratios | Example of clinical situation | |
| (1) | RTMF and RPAF = 1 ± 5% | Normal |
| (2) | RTMF > 1.5 (and RPAF > 1.5) | Atrial left-to-right shunt |
| (3) | RTMF < 1.5 (and RPAF < 1.5) | Atrial right-to-left shunt |
| (4) | RTMF normal and RPAF > 1.5 | Pulmonary regurgitation |
| (5) | RTMF normal and RPAF < 1.5 | Aortic regurgitation |
Indeed, the comparison of these measurements with the unity may identify typical pathological features, helping the establishment of the diagnostic. Each one of the above-mentioned clinical situations can be interpreted as follows:
- The amount of blood that is sent to the pulmonary circulation through the tricuspid and the pulmonary valve is similar to the amount of blood sent to the systemic circulation through the mitral and the aortic valve.
- If RTMF > 1 ±5% (and so probably RPAF > 1 ± 5%), then the flow across the tricuspid valve is significantly higher than the flow across the mitral valve, which means that the right ventricle receives more blood than the left one. If the atria receives the same amount of blood we can deduce then that there is a left to right shunt. This parameter is crucial when evaluating atrial septum defects for example.
- If RTMF < 1 ±5% (and so probably RPAF < 1 ± 5%), the phenomenon is inverted and in this case we can deduce that the shunt is right to left.
- RTMF is normal but RPAF > 1 ± 5%. In this case there is more blood that passes through the pulmonary valve than through the aortic valve. One can deduce thus that there is a pulmonary regurgitation. This situation is found in post-operative patients who suffered from tetralogy of Fallot for instance.
- RTMF is normal but RPAF < 1 ± 5%. This case is opposite to the previous situation, there is an aortic regurgitation. This phenomenon can be seen for instance in some hypertrophied cardiomyopathies where the pathological and thick septum shifts the aortic valve which, as a result, will not close accordingly.
Finally, RPAF ratio may also help the expert to detect shunts ant to assess them as:
- trivial: if RPAF is greater than 1.05 but lower than 1.5,
- mild: if RPAF is greater than 1.5 but lower than 2,
- significant: if RPAF is greater than 2 but lower than 3,
- severe: if RPAF is greater than 3.
Second step: volume and contractility comparisons
Though comparing the flow across the various cardiac valves allows the cardiologists to detect some abnormal functions of the heart, other parameters may be useful to diagnose diseases that do not affect the blood circulation directly. In those cases, volumetric parameters may highlight atypical features of the myocardium and help the expert to identify cardiomyopathies or severe right-ventricle overload. For example,
- if the flows are normal and the volume is significantly large (+2 standard deviations), then the contractility of the muscle is abnormal. Actually, ventricles with poor contractility need to receive more blood in order to eject the same quantity of blood in the circulation system,
- if the flows are abnormal and the volume is significantly large, then a possible cause of the dilatation of the ventricle may be an abnormal blood circulation. This is typically the case in right-ventricular overload and in tetralogy of Fallot where the ventricle is dilated because of the increased pre-load. In atrial septal defect for example the abnormal flow is at the level of the atrio-ventricular valves whereas in post-operative tetralogy of Fallot the regurgitations of the pulmonary valve lead to a dilatation of the ventricle.
5. Conclusion
The main idea of knowledge-based models is to rely exclusively on standardised measurements that are acquired during clinical routine and to organise them into an in-silico algorithm developed from the methods that are used by the cardiologists.
A selection of relevant measurements that best describe the heart anatomy in children is presented in this document. The aim is to select and define wise clinical parameters that allow objective discrimination between the various pathologies. So far, cavity volumes, contractilities and flow measurements have been selected but other measurements might be added.
We have then proposed a preliminary knowledge-based model that allows the identification of typical clinical situations by analysing the patient's parameters. Moreover, such a method may also be applied to assess the severity of a disease by comparing the parameters to a standard scale.
However, the model might be improved by adding for instance more discriminant measurements to identify subtle pathological behaviours. This drives us to consider interesting challenges, such as the adjustment of the available imaging techniques to the specific case of paediatric diseases and to the specificities of the evaluation of the right-ventricle. Similarly, the decision tests may be refined in order to take into account all the possible clinical cases, from the most evident to the most subtle.